2 * Monkey's Audio lossless audio decoder
3 * Copyright (c) 2007 Benjamin Zores <ben@geexbox.org>
4 * based upon libdemac from Dave Chapman.
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25 #include "libavutil/avassert.h"
26 #include "libavutil/channel_layout.h"
27 #include "libavutil/crc.h"
28 #include "libavutil/opt.h"
29 #include "lossless_audiodsp.h"
32 #include "bytestream.h"
39 * Monkey's Audio lossless audio decoder
42 #define MAX_CHANNELS 2
43 #define MAX_BYTESPERSAMPLE 3
45 #define APE_FRAMECODE_MONO_SILENCE 1
46 #define APE_FRAMECODE_STEREO_SILENCE 3
47 #define APE_FRAMECODE_PSEUDO_STEREO 4
49 #define HISTORY_SIZE 512
50 #define PREDICTOR_ORDER 8
51 /** Total size of all predictor histories */
52 #define PREDICTOR_SIZE 50
54 #define YDELAYA (18 + PREDICTOR_ORDER*4)
55 #define YDELAYB (18 + PREDICTOR_ORDER*3)
56 #define XDELAYA (18 + PREDICTOR_ORDER*2)
57 #define XDELAYB (18 + PREDICTOR_ORDER)
59 #define YADAPTCOEFFSA 18
60 #define XADAPTCOEFFSA 14
61 #define YADAPTCOEFFSB 10
62 #define XADAPTCOEFFSB 5
65 * Possible compression levels
68 enum APECompressionLevel {
69 COMPRESSION_LEVEL_FAST = 1000,
70 COMPRESSION_LEVEL_NORMAL = 2000,
71 COMPRESSION_LEVEL_HIGH = 3000,
72 COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
73 COMPRESSION_LEVEL_INSANE = 5000
77 #define APE_FILTER_LEVELS 3
79 /** Filter orders depending on compression level */
80 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
88 /** Filter fraction bits depending on compression level */
89 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
98 /** Filters applied to the decoded data */
99 typedef struct APEFilter {
100 int16_t *coeffs; ///< actual coefficients used in filtering
101 int16_t *adaptcoeffs; ///< adaptive filter coefficients used for correcting of actual filter coefficients
102 int16_t *historybuffer; ///< filter memory
103 int16_t *delay; ///< filtered values
108 typedef struct APERice {
113 typedef struct APERangecoder {
114 uint32_t low; ///< low end of interval
115 uint32_t range; ///< length of interval
116 uint32_t help; ///< bytes_to_follow resp. intermediate value
117 unsigned int buffer; ///< buffer for input/output
120 /** Filter histories */
121 typedef struct APEPredictor {
129 uint32_t coeffsA[2][4]; ///< adaption coefficients
130 uint32_t coeffsB[2][5]; ///< adaption coefficients
131 int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
133 unsigned int sample_pos;
136 /** Decoder context */
137 typedef struct APEContext {
138 AVClass *class; ///< class for AVOptions
139 AVCodecContext *avctx;
140 BswapDSPContext bdsp;
141 LLAudDSPContext adsp;
143 int samples; ///< samples left to decode in current frame
146 int fileversion; ///< codec version, very important in decoding process
147 int compression_level; ///< compression levels
148 int fset; ///< which filter set to use (calculated from compression level)
149 int flags; ///< global decoder flags
151 uint32_t CRC; ///< signalled frame CRC
152 uint32_t CRC_state; ///< accumulated CRC
153 int frameflags; ///< frame flags
154 APEPredictor predictor; ///< predictor used for final reconstruction
156 int32_t *decoded_buffer;
158 int32_t *decoded[MAX_CHANNELS]; ///< decoded data for each channel
159 int blocks_per_loop; ///< maximum number of samples to decode for each call
161 int16_t* filterbuf[APE_FILTER_LEVELS]; ///< filter memory
163 APERangecoder rc; ///< rangecoder used to decode actual values
164 APERice riceX; ///< rice code parameters for the second channel
165 APERice riceY; ///< rice code parameters for the first channel
166 APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
169 uint8_t *data; ///< current frame data
170 uint8_t *data_end; ///< frame data end
171 int data_size; ///< frame data allocated size
172 const uint8_t *ptr; ///< current position in frame data
176 void (*entropy_decode_mono)(struct APEContext *ctx, int blockstodecode);
177 void (*entropy_decode_stereo)(struct APEContext *ctx, int blockstodecode);
178 void (*predictor_decode_mono)(struct APEContext *ctx, int count);
179 void (*predictor_decode_stereo)(struct APEContext *ctx, int count);
182 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
183 int32_t *decoded1, int count);
185 static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode);
186 static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode);
187 static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode);
188 static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode);
189 static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode);
190 static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode);
191 static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode);
192 static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode);
193 static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode);
195 static void predictor_decode_mono_3800(APEContext *ctx, int count);
196 static void predictor_decode_stereo_3800(APEContext *ctx, int count);
197 static void predictor_decode_mono_3930(APEContext *ctx, int count);
198 static void predictor_decode_stereo_3930(APEContext *ctx, int count);
199 static void predictor_decode_mono_3950(APEContext *ctx, int count);
200 static void predictor_decode_stereo_3950(APEContext *ctx, int count);
202 static av_cold int ape_decode_close(AVCodecContext *avctx)
204 APEContext *s = avctx->priv_data;
207 for (i = 0; i < APE_FILTER_LEVELS; i++)
208 av_freep(&s->filterbuf[i]);
210 av_freep(&s->decoded_buffer);
212 s->decoded_size = s->data_size = 0;
217 static av_cold int ape_decode_init(AVCodecContext *avctx)
219 APEContext *s = avctx->priv_data;
222 if (avctx->extradata_size != 6) {
223 av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
224 return AVERROR(EINVAL);
226 if (avctx->channels > 2) {
227 av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
228 return AVERROR(EINVAL);
230 s->bps = avctx->bits_per_coded_sample;
233 avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
236 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
239 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
242 avpriv_request_sample(avctx,
243 "%d bits per coded sample", s->bps);
244 return AVERROR_PATCHWELCOME;
247 s->channels = avctx->channels;
248 s->fileversion = AV_RL16(avctx->extradata);
249 s->compression_level = AV_RL16(avctx->extradata + 2);
250 s->flags = AV_RL16(avctx->extradata + 4);
252 av_log(avctx, AV_LOG_VERBOSE, "Compression Level: %d - Flags: %d\n",
253 s->compression_level, s->flags);
254 if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE ||
255 !s->compression_level ||
256 (s->fileversion < 3930 && s->compression_level == COMPRESSION_LEVEL_INSANE)) {
257 av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
258 s->compression_level);
259 return AVERROR_INVALIDDATA;
261 s->fset = s->compression_level / 1000 - 1;
262 for (i = 0; i < APE_FILTER_LEVELS; i++) {
263 if (!ape_filter_orders[s->fset][i])
265 if (!(s->filterbuf[i] = av_malloc((ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4)))
266 return AVERROR(ENOMEM);
269 if (s->fileversion < 3860) {
270 s->entropy_decode_mono = entropy_decode_mono_0000;
271 s->entropy_decode_stereo = entropy_decode_stereo_0000;
272 } else if (s->fileversion < 3900) {
273 s->entropy_decode_mono = entropy_decode_mono_3860;
274 s->entropy_decode_stereo = entropy_decode_stereo_3860;
275 } else if (s->fileversion < 3930) {
276 s->entropy_decode_mono = entropy_decode_mono_3900;
277 s->entropy_decode_stereo = entropy_decode_stereo_3900;
278 } else if (s->fileversion < 3990) {
279 s->entropy_decode_mono = entropy_decode_mono_3900;
280 s->entropy_decode_stereo = entropy_decode_stereo_3930;
282 s->entropy_decode_mono = entropy_decode_mono_3990;
283 s->entropy_decode_stereo = entropy_decode_stereo_3990;
286 if (s->fileversion < 3930) {
287 s->predictor_decode_mono = predictor_decode_mono_3800;
288 s->predictor_decode_stereo = predictor_decode_stereo_3800;
289 } else if (s->fileversion < 3950) {
290 s->predictor_decode_mono = predictor_decode_mono_3930;
291 s->predictor_decode_stereo = predictor_decode_stereo_3930;
293 s->predictor_decode_mono = predictor_decode_mono_3950;
294 s->predictor_decode_stereo = predictor_decode_stereo_3950;
297 ff_bswapdsp_init(&s->bdsp);
298 ff_llauddsp_init(&s->adsp);
299 avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
305 * @name APE range decoding functions
310 #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1))
311 #define SHIFT_BITS (CODE_BITS - 9)
312 #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1)
313 #define BOTTOM_VALUE (TOP_VALUE >> 8)
315 /** Start the decoder */
316 static inline void range_start_decoding(APEContext *ctx)
318 ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
319 ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS);
320 ctx->rc.range = (uint32_t) 1 << EXTRA_BITS;
323 /** Perform normalization */
324 static inline void range_dec_normalize(APEContext *ctx)
326 while (ctx->rc.range <= BOTTOM_VALUE) {
327 ctx->rc.buffer <<= 8;
328 if(ctx->ptr < ctx->data_end) {
329 ctx->rc.buffer += *ctx->ptr;
334 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF);
340 * Calculate cumulative frequency for next symbol. Does NO update!
341 * @param ctx decoder context
342 * @param tot_f is the total frequency or (code_value)1<<shift
343 * @return the cumulative frequency
345 static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
347 range_dec_normalize(ctx);
348 ctx->rc.help = ctx->rc.range / tot_f;
349 return ctx->rc.low / ctx->rc.help;
353 * Decode value with given size in bits
354 * @param ctx decoder context
355 * @param shift number of bits to decode
357 static inline int range_decode_culshift(APEContext *ctx, int shift)
359 range_dec_normalize(ctx);
360 ctx->rc.help = ctx->rc.range >> shift;
361 return ctx->rc.low / ctx->rc.help;
366 * Update decoding state
367 * @param ctx decoder context
368 * @param sy_f the interval length (frequency of the symbol)
369 * @param lt_f the lower end (frequency sum of < symbols)
371 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
373 ctx->rc.low -= ctx->rc.help * lt_f;
374 ctx->rc.range = ctx->rc.help * sy_f;
377 /** Decode n bits (n <= 16) without modelling */
378 static inline int range_decode_bits(APEContext *ctx, int n)
380 int sym = range_decode_culshift(ctx, n);
381 range_decode_update(ctx, 1, sym);
386 #define MODEL_ELEMENTS 64
389 * Fixed probabilities for symbols in Monkey Audio version 3.97
391 static const uint16_t counts_3970[22] = {
392 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
393 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
394 65450, 65469, 65480, 65487, 65491, 65493,
398 * Probability ranges for symbols in Monkey Audio version 3.97
400 static const uint16_t counts_diff_3970[21] = {
401 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
402 1104, 677, 415, 248, 150, 89, 54, 31,
407 * Fixed probabilities for symbols in Monkey Audio version 3.98
409 static const uint16_t counts_3980[22] = {
410 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
411 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
412 65485, 65488, 65490, 65491, 65492, 65493,
416 * Probability ranges for symbols in Monkey Audio version 3.98
418 static const uint16_t counts_diff_3980[21] = {
419 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
420 261, 119, 65, 31, 19, 10, 6, 3,
426 * @param ctx decoder context
427 * @param counts probability range start position
428 * @param counts_diff probability range widths
430 static inline int range_get_symbol(APEContext *ctx,
431 const uint16_t counts[],
432 const uint16_t counts_diff[])
436 cf = range_decode_culshift(ctx, 16);
439 symbol= cf - 65535 + 63;
440 range_decode_update(ctx, 1, cf);
445 /* figure out the symbol inefficiently; a binary search would be much better */
446 for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
448 range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
452 /** @} */ // group rangecoder
454 static inline void update_rice(APERice *rice, unsigned int x)
456 int lim = rice->k ? (1 << (rice->k + 4)) : 0;
457 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
459 if (rice->ksum < lim)
461 else if (rice->ksum >= (1 << (rice->k + 5)) && rice->k < 24)
465 static inline int get_rice_ook(GetBitContext *gb, int k)
469 x = get_unary(gb, 1, get_bits_left(gb));
472 x = (x << k) | get_bits(gb, k);
477 static inline int ape_decode_value_3860(APEContext *ctx, GetBitContext *gb,
480 unsigned int x, overflow;
482 overflow = get_unary(gb, 1, get_bits_left(gb));
484 if (ctx->fileversion > 3880) {
485 while (overflow >= 16) {
493 else if(rice->k <= MIN_CACHE_BITS) {
494 x = (overflow << rice->k) + get_bits(gb, rice->k);
496 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %"PRIu32"\n", rice->k);
498 return AVERROR_INVALIDDATA;
500 rice->ksum += x - (rice->ksum + 8 >> 4);
501 if (rice->ksum < (rice->k ? 1 << (rice->k + 4) : 0))
503 else if (rice->ksum >= (1 << (rice->k + 5)) && rice->k < 24)
506 /* Convert to signed */
507 return ((x >> 1) ^ ((x & 1) - 1)) + 1;
510 static inline int ape_decode_value_3900(APEContext *ctx, APERice *rice)
512 unsigned int x, overflow;
515 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
517 if (overflow == (MODEL_ELEMENTS - 1)) {
518 tmpk = range_decode_bits(ctx, 5);
521 tmpk = (rice->k < 1) ? 0 : rice->k - 1;
523 if (tmpk <= 16 || ctx->fileversion < 3910) {
525 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
526 return AVERROR_INVALIDDATA;
528 x = range_decode_bits(ctx, tmpk);
529 } else if (tmpk <= 31) {
530 x = range_decode_bits(ctx, 16);
531 x |= (range_decode_bits(ctx, tmpk - 16) << 16);
533 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
534 return AVERROR_INVALIDDATA;
536 x += overflow << tmpk;
538 update_rice(rice, x);
540 /* Convert to signed */
541 return ((x >> 1) ^ ((x & 1) - 1)) + 1;
544 static inline int ape_decode_value_3990(APEContext *ctx, APERice *rice)
546 unsigned int x, overflow;
549 pivot = rice->ksum >> 5;
553 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
555 if (overflow == (MODEL_ELEMENTS - 1)) {
556 overflow = (unsigned)range_decode_bits(ctx, 16) << 16;
557 overflow |= range_decode_bits(ctx, 16);
560 if (pivot < 0x10000) {
561 base = range_decode_culfreq(ctx, pivot);
562 range_decode_update(ctx, 1, base);
564 int base_hi = pivot, base_lo;
567 while (base_hi & ~0xFFFF) {
571 base_hi = range_decode_culfreq(ctx, base_hi + 1);
572 range_decode_update(ctx, 1, base_hi);
573 base_lo = range_decode_culfreq(ctx, 1 << bbits);
574 range_decode_update(ctx, 1, base_lo);
576 base = (base_hi << bbits) + base_lo;
579 x = base + overflow * pivot;
581 update_rice(rice, x);
583 /* Convert to signed */
584 return ((x >> 1) ^ ((x & 1) - 1)) + 1;
587 static int get_k(int ksum)
589 return av_log2(ksum) + !!ksum;
592 static void decode_array_0000(APEContext *ctx, GetBitContext *gb,
593 int32_t *out, APERice *rice, int blockstodecode)
596 unsigned ksummax, ksummin;
599 for (i = 0; i < FFMIN(blockstodecode, 5); i++) {
600 out[i] = get_rice_ook(&ctx->gb, 10);
601 rice->ksum += out[i];
604 if (blockstodecode <= 5)
607 rice->k = get_k(rice->ksum / 10);
610 for (; i < FFMIN(blockstodecode, 64); i++) {
611 out[i] = get_rice_ook(&ctx->gb, rice->k);
612 rice->ksum += out[i];
613 rice->k = get_k(rice->ksum / ((i + 1) * 2));
618 if (blockstodecode <= 64)
621 rice->k = get_k(rice->ksum >> 7);
622 ksummax = 1 << rice->k + 7;
623 ksummin = rice->k ? (1 << rice->k + 6) : 0;
624 for (; i < blockstodecode; i++) {
625 if (get_bits_left(&ctx->gb) < 1) {
629 out[i] = get_rice_ook(&ctx->gb, rice->k);
630 rice->ksum += out[i] - (unsigned)out[i - 64];
631 while (rice->ksum < ksummin) {
633 ksummin = rice->k ? ksummin >> 1 : 0;
636 while (rice->ksum >= ksummax) {
641 ksummin = ksummin ? ksummin << 1 : 128;
646 for (i = 0; i < blockstodecode; i++)
647 out[i] = ((out[i] >> 1) ^ ((out[i] & 1) - 1)) + 1;
650 static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode)
652 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
656 static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode)
658 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
660 decode_array_0000(ctx, &ctx->gb, ctx->decoded[1], &ctx->riceX,
664 static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode)
666 int32_t *decoded0 = ctx->decoded[0];
668 while (blockstodecode--)
669 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
672 static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode)
674 int32_t *decoded0 = ctx->decoded[0];
675 int32_t *decoded1 = ctx->decoded[1];
676 int blocks = blockstodecode;
678 while (blockstodecode--)
679 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
681 *decoded1++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceX);
684 static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode)
686 int32_t *decoded0 = ctx->decoded[0];
688 while (blockstodecode--)
689 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
692 static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode)
694 int32_t *decoded0 = ctx->decoded[0];
695 int32_t *decoded1 = ctx->decoded[1];
696 int blocks = blockstodecode;
698 while (blockstodecode--)
699 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
700 range_dec_normalize(ctx);
701 // because of some implementation peculiarities we need to backpedal here
703 range_start_decoding(ctx);
705 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
708 static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode)
710 int32_t *decoded0 = ctx->decoded[0];
711 int32_t *decoded1 = ctx->decoded[1];
713 while (blockstodecode--) {
714 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
715 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
719 static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode)
721 int32_t *decoded0 = ctx->decoded[0];
723 while (blockstodecode--)
724 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
727 static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode)
729 int32_t *decoded0 = ctx->decoded[0];
730 int32_t *decoded1 = ctx->decoded[1];
732 while (blockstodecode--) {
733 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
734 *decoded1++ = ape_decode_value_3990(ctx, &ctx->riceX);
738 static int init_entropy_decoder(APEContext *ctx)
741 if (ctx->fileversion >= 3900) {
742 if (ctx->data_end - ctx->ptr < 6)
743 return AVERROR_INVALIDDATA;
744 ctx->CRC = bytestream_get_be32(&ctx->ptr);
746 ctx->CRC = get_bits_long(&ctx->gb, 32);
749 /* Read the frame flags if they exist */
751 ctx->CRC_state = UINT32_MAX;
752 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
753 ctx->CRC &= ~0x80000000;
755 if (ctx->data_end - ctx->ptr < 6)
756 return AVERROR_INVALIDDATA;
757 ctx->frameflags = bytestream_get_be32(&ctx->ptr);
760 /* Initialize the rice structs */
762 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
764 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
766 if (ctx->fileversion >= 3900) {
767 /* The first 8 bits of input are ignored. */
770 range_start_decoding(ctx);
776 static const int32_t initial_coeffs_fast_3320[1] = {
780 static const int32_t initial_coeffs_a_3800[3] = {
784 static const int32_t initial_coeffs_b_3800[2] = {
788 static const int32_t initial_coeffs_3930[4] = {
792 static void init_predictor_decoder(APEContext *ctx)
794 APEPredictor *p = &ctx->predictor;
796 /* Zero the history buffers */
797 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
798 p->buf = p->historybuffer;
800 /* Initialize and zero the coefficients */
801 if (ctx->fileversion < 3930) {
802 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
803 memcpy(p->coeffsA[0], initial_coeffs_fast_3320,
804 sizeof(initial_coeffs_fast_3320));
805 memcpy(p->coeffsA[1], initial_coeffs_fast_3320,
806 sizeof(initial_coeffs_fast_3320));
808 memcpy(p->coeffsA[0], initial_coeffs_a_3800,
809 sizeof(initial_coeffs_a_3800));
810 memcpy(p->coeffsA[1], initial_coeffs_a_3800,
811 sizeof(initial_coeffs_a_3800));
814 memcpy(p->coeffsA[0], initial_coeffs_3930, sizeof(initial_coeffs_3930));
815 memcpy(p->coeffsA[1], initial_coeffs_3930, sizeof(initial_coeffs_3930));
817 memset(p->coeffsB, 0, sizeof(p->coeffsB));
818 if (ctx->fileversion < 3930) {
819 memcpy(p->coeffsB[0], initial_coeffs_b_3800,
820 sizeof(initial_coeffs_b_3800));
821 memcpy(p->coeffsB[1], initial_coeffs_b_3800,
822 sizeof(initial_coeffs_b_3800));
825 p->filterA[0] = p->filterA[1] = 0;
826 p->filterB[0] = p->filterB[1] = 0;
827 p->lastA[0] = p->lastA[1] = 0;
832 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
833 static inline int APESIGN(int32_t x) {
834 return (x < 0) - (x > 0);
837 static av_always_inline int filter_fast_3320(APEPredictor *p,
838 const int decoded, const int filter,
843 p->buf[delayA] = p->lastA[filter];
844 if (p->sample_pos < 3) {
845 p->lastA[filter] = decoded;
846 p->filterA[filter] = decoded;
850 predictionA = p->buf[delayA] * 2U - p->buf[delayA - 1];
851 p->lastA[filter] = decoded + ((int32_t)(predictionA * p->coeffsA[filter][0]) >> 9);
853 if ((decoded ^ predictionA) > 0)
854 p->coeffsA[filter][0]++;
856 p->coeffsA[filter][0]--;
858 p->filterA[filter] += (unsigned)p->lastA[filter];
860 return p->filterA[filter];
863 static av_always_inline int filter_3800(APEPredictor *p,
864 const unsigned decoded, const int filter,
865 const int delayA, const int delayB,
866 const int start, const int shift)
868 int32_t predictionA, predictionB, sign;
869 int32_t d0, d1, d2, d3, d4;
871 p->buf[delayA] = p->lastA[filter];
872 p->buf[delayB] = p->filterB[filter];
873 if (p->sample_pos < start) {
874 predictionA = decoded + p->filterA[filter];
875 p->lastA[filter] = decoded;
876 p->filterB[filter] = decoded;
877 p->filterA[filter] = predictionA;
881 d1 = (p->buf[delayA] - p->buf[delayA - 1]) * 2U;
882 d0 = p->buf[delayA] + ((p->buf[delayA - 2] - p->buf[delayA - 1]) * 8U);
883 d3 = p->buf[delayB] * 2U - p->buf[delayB - 1];
886 predictionA = d0 * p->coeffsA[filter][0] +
887 d1 * p->coeffsA[filter][1] +
888 d2 * p->coeffsA[filter][2];
890 sign = APESIGN(decoded);
891 p->coeffsA[filter][0] += (((d0 >> 30) & 2) - 1) * sign;
892 p->coeffsA[filter][1] += (((d1 >> 28) & 8) - 4) * sign;
893 p->coeffsA[filter][2] += (((d2 >> 28) & 8) - 4) * sign;
895 predictionB = d3 * p->coeffsB[filter][0] -
896 d4 * p->coeffsB[filter][1];
897 p->lastA[filter] = decoded + (predictionA >> 11);
898 sign = APESIGN(p->lastA[filter]);
899 p->coeffsB[filter][0] += (((d3 >> 29) & 4) - 2) * sign;
900 p->coeffsB[filter][1] -= (((d4 >> 30) & 2) - 1) * sign;
902 p->filterB[filter] = p->lastA[filter] + (predictionB >> shift);
903 p->filterA[filter] = p->filterB[filter] + (unsigned)((int)(p->filterA[filter] * 31U) >> 5);
905 return p->filterA[filter];
908 static void long_filter_high_3800(int32_t *buffer, int order, int shift, int length)
911 int32_t dotprod, sign;
912 int32_t coeffs[256], delay[256];
917 memset(coeffs, 0, order * sizeof(*coeffs));
918 for (i = 0; i < order; i++)
919 delay[i] = buffer[i];
920 for (i = order; i < length; i++) {
922 sign = APESIGN(buffer[i]);
923 for (j = 0; j < order; j++) {
924 dotprod += delay[j] * (unsigned)coeffs[j];
925 coeffs[j] += ((delay[j] >> 31) | 1) * sign;
927 buffer[i] -= dotprod >> shift;
928 for (j = 0; j < order - 1; j++)
929 delay[j] = delay[j + 1];
930 delay[order - 1] = buffer[i];
934 static void long_filter_ehigh_3830(int32_t *buffer, int length)
937 int32_t dotprod, sign;
938 int32_t delay[8] = { 0 };
939 uint32_t coeffs[8] = { 0 };
941 for (i = 0; i < length; i++) {
943 sign = APESIGN(buffer[i]);
944 for (j = 7; j >= 0; j--) {
945 dotprod += delay[j] * coeffs[j];
946 coeffs[j] += ((delay[j] >> 31) | 1) * sign;
948 for (j = 7; j > 0; j--)
949 delay[j] = delay[j - 1];
950 delay[0] = buffer[i];
951 buffer[i] -= dotprod >> 9;
955 static void predictor_decode_stereo_3800(APEContext *ctx, int count)
957 APEPredictor *p = &ctx->predictor;
958 int32_t *decoded0 = ctx->decoded[0];
959 int32_t *decoded1 = ctx->decoded[1];
960 int start = 4, shift = 10;
962 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
964 long_filter_high_3800(decoded0, 16, 9, count);
965 long_filter_high_3800(decoded1, 16, 9, count);
966 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
967 int order = 128, shift2 = 11;
969 if (ctx->fileversion >= 3830) {
973 long_filter_ehigh_3830(decoded0 + order, count - order);
974 long_filter_ehigh_3830(decoded1 + order, count - order);
977 long_filter_high_3800(decoded0, order, shift2, count);
978 long_filter_high_3800(decoded1, order, shift2, count);
982 int X = *decoded0, Y = *decoded1;
983 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
984 *decoded0 = filter_fast_3320(p, Y, 0, YDELAYA);
986 *decoded1 = filter_fast_3320(p, X, 1, XDELAYA);
989 *decoded0 = filter_3800(p, Y, 0, YDELAYA, YDELAYB,
992 *decoded1 = filter_3800(p, X, 1, XDELAYA, XDELAYB,
1001 /* Have we filled the history buffer? */
1002 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1003 memmove(p->historybuffer, p->buf,
1004 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1005 p->buf = p->historybuffer;
1010 static void predictor_decode_mono_3800(APEContext *ctx, int count)
1012 APEPredictor *p = &ctx->predictor;
1013 int32_t *decoded0 = ctx->decoded[0];
1014 int start = 4, shift = 10;
1016 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
1018 long_filter_high_3800(decoded0, 16, 9, count);
1019 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
1020 int order = 128, shift2 = 11;
1022 if (ctx->fileversion >= 3830) {
1026 long_filter_ehigh_3830(decoded0 + order, count - order);
1029 long_filter_high_3800(decoded0, order, shift2, count);
1033 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
1034 *decoded0 = filter_fast_3320(p, *decoded0, 0, YDELAYA);
1037 *decoded0 = filter_3800(p, *decoded0, 0, YDELAYA, YDELAYB,
1046 /* Have we filled the history buffer? */
1047 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1048 memmove(p->historybuffer, p->buf,
1049 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1050 p->buf = p->historybuffer;
1055 static av_always_inline int predictor_update_3930(APEPredictor *p,
1056 const int decoded, const int filter,
1059 int32_t predictionA, sign;
1060 int32_t d0, d1, d2, d3;
1062 p->buf[delayA] = p->lastA[filter];
1063 d0 = p->buf[delayA ];
1064 d1 = p->buf[delayA ] - p->buf[delayA - 1];
1065 d2 = p->buf[delayA - 1] - p->buf[delayA - 2];
1066 d3 = p->buf[delayA - 2] - p->buf[delayA - 3];
1068 predictionA = d0 * p->coeffsA[filter][0] +
1069 d1 * p->coeffsA[filter][1] +
1070 d2 * p->coeffsA[filter][2] +
1071 d3 * p->coeffsA[filter][3];
1073 p->lastA[filter] = decoded + (predictionA >> 9);
1074 p->filterA[filter] = p->lastA[filter] + ((int)(p->filterA[filter] * 31U) >> 5);
1076 sign = APESIGN(decoded);
1077 p->coeffsA[filter][0] += ((d0 < 0) * 2 - 1) * sign;
1078 p->coeffsA[filter][1] += ((d1 < 0) * 2 - 1) * sign;
1079 p->coeffsA[filter][2] += ((d2 < 0) * 2 - 1) * sign;
1080 p->coeffsA[filter][3] += ((d3 < 0) * 2 - 1) * sign;
1082 return p->filterA[filter];
1085 static void predictor_decode_stereo_3930(APEContext *ctx, int count)
1087 APEPredictor *p = &ctx->predictor;
1088 int32_t *decoded0 = ctx->decoded[0];
1089 int32_t *decoded1 = ctx->decoded[1];
1091 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1095 int Y = *decoded1, X = *decoded0;
1096 *decoded0 = predictor_update_3930(p, Y, 0, YDELAYA);
1098 *decoded1 = predictor_update_3930(p, X, 1, XDELAYA);
1104 /* Have we filled the history buffer? */
1105 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1106 memmove(p->historybuffer, p->buf,
1107 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1108 p->buf = p->historybuffer;
1113 static void predictor_decode_mono_3930(APEContext *ctx, int count)
1115 APEPredictor *p = &ctx->predictor;
1116 int32_t *decoded0 = ctx->decoded[0];
1118 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1121 *decoded0 = predictor_update_3930(p, *decoded0, 0, YDELAYA);
1126 /* Have we filled the history buffer? */
1127 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1128 memmove(p->historybuffer, p->buf,
1129 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1130 p->buf = p->historybuffer;
1135 static av_always_inline int predictor_update_filter(APEPredictor *p,
1136 const int decoded, const int filter,
1137 const int delayA, const int delayB,
1138 const int adaptA, const int adaptB)
1140 int32_t predictionA, predictionB, sign;
1142 p->buf[delayA] = p->lastA[filter];
1143 p->buf[adaptA] = APESIGN(p->buf[delayA]);
1144 p->buf[delayA - 1] = p->buf[delayA] - (unsigned)p->buf[delayA - 1];
1145 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
1147 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
1148 p->buf[delayA - 1] * p->coeffsA[filter][1] +
1149 p->buf[delayA - 2] * p->coeffsA[filter][2] +
1150 p->buf[delayA - 3] * p->coeffsA[filter][3];
1152 /* Apply a scaled first-order filter compression */
1153 p->buf[delayB] = p->filterA[filter ^ 1] - ((int)(p->filterB[filter] * 31U) >> 5);
1154 p->buf[adaptB] = APESIGN(p->buf[delayB]);
1155 p->buf[delayB - 1] = p->buf[delayB] - (unsigned)p->buf[delayB - 1];
1156 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
1157 p->filterB[filter] = p->filterA[filter ^ 1];
1159 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
1160 p->buf[delayB - 1] * p->coeffsB[filter][1] +
1161 p->buf[delayB - 2] * p->coeffsB[filter][2] +
1162 p->buf[delayB - 3] * p->coeffsB[filter][3] +
1163 p->buf[delayB - 4] * p->coeffsB[filter][4];
1165 p->lastA[filter] = decoded + ((int)((unsigned)predictionA + (predictionB >> 1)) >> 10);
1166 p->filterA[filter] = p->lastA[filter] + ((int)(p->filterA[filter] * 31U) >> 5);
1168 sign = APESIGN(decoded);
1169 p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
1170 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
1171 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
1172 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
1173 p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
1174 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
1175 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
1176 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
1177 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
1179 return p->filterA[filter];
1182 static void predictor_decode_stereo_3950(APEContext *ctx, int count)
1184 APEPredictor *p = &ctx->predictor;
1185 int32_t *decoded0 = ctx->decoded[0];
1186 int32_t *decoded1 = ctx->decoded[1];
1188 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1192 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
1193 YADAPTCOEFFSA, YADAPTCOEFFSB);
1195 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
1196 XADAPTCOEFFSA, XADAPTCOEFFSB);
1202 /* Have we filled the history buffer? */
1203 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1204 memmove(p->historybuffer, p->buf,
1205 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1206 p->buf = p->historybuffer;
1211 static void predictor_decode_mono_3950(APEContext *ctx, int count)
1213 APEPredictor *p = &ctx->predictor;
1214 int32_t *decoded0 = ctx->decoded[0];
1215 int32_t predictionA, currentA, A, sign;
1217 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1219 currentA = p->lastA[0];
1224 p->buf[YDELAYA] = currentA;
1225 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - (unsigned)p->buf[YDELAYA - 1];
1227 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
1228 p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
1229 p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
1230 p->buf[YDELAYA - 3] * p->coeffsA[0][3];
1232 currentA = A + (unsigned)(predictionA >> 10);
1234 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
1235 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
1238 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
1239 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
1240 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
1241 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
1245 /* Have we filled the history buffer? */
1246 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1247 memmove(p->historybuffer, p->buf,
1248 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1249 p->buf = p->historybuffer;
1252 p->filterA[0] = currentA + (unsigned)((int)(p->filterA[0] * 31U) >> 5);
1253 *(decoded0++) = p->filterA[0];
1256 p->lastA[0] = currentA;
1259 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
1262 f->historybuffer = buf + order;
1263 f->delay = f->historybuffer + order * 2;
1264 f->adaptcoeffs = f->historybuffer + order;
1266 memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
1267 memset(f->coeffs, 0, order * sizeof(*f->coeffs));
1271 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
1273 do_init_filter(&f[0], buf, order);
1274 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
1277 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
1278 int32_t *data, int count, int order, int fracbits)
1284 /* round fixedpoint scalar product */
1285 res = ctx->adsp.scalarproduct_and_madd_int16(f->coeffs,
1287 f->adaptcoeffs - order,
1288 order, APESIGN(*data));
1289 res = (int)(res + (1U << (fracbits - 1))) >> fracbits;
1290 res += (unsigned)*data;
1293 /* Update the output history */
1294 *f->delay++ = av_clip_int16(res);
1296 if (version < 3980) {
1297 /* Version ??? to < 3.98 files (untested) */
1298 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
1299 f->adaptcoeffs[-4] >>= 1;
1300 f->adaptcoeffs[-8] >>= 1;
1302 /* Version 3.98 and later files */
1304 /* Update the adaption coefficients */
1305 absres = res < 0 ? -(unsigned)res : res;
1307 *f->adaptcoeffs = APESIGN(res) *
1308 (8 << ((absres > f->avg * 3) + (absres > f->avg * 4 / 3)));
1309 /* equivalent to the following code
1310 if (absres <= f->avg * 4 / 3)
1311 *f->adaptcoeffs = APESIGN(res) * 8;
1312 else if (absres <= f->avg * 3)
1313 *f->adaptcoeffs = APESIGN(res) * 16;
1315 *f->adaptcoeffs = APESIGN(res) * 32;
1318 *f->adaptcoeffs = 0;
1320 f->avg += (int)(absres - (unsigned)f->avg) / 16;
1322 f->adaptcoeffs[-1] >>= 1;
1323 f->adaptcoeffs[-2] >>= 1;
1324 f->adaptcoeffs[-8] >>= 1;
1329 /* Have we filled the history buffer? */
1330 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
1331 memmove(f->historybuffer, f->delay - (order * 2),
1332 (order * 2) * sizeof(*f->historybuffer));
1333 f->delay = f->historybuffer + order * 2;
1334 f->adaptcoeffs = f->historybuffer + order;
1339 static void apply_filter(APEContext *ctx, APEFilter *f,
1340 int32_t *data0, int32_t *data1,
1341 int count, int order, int fracbits)
1343 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
1345 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
1348 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
1349 int32_t *decoded1, int count)
1353 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1354 if (!ape_filter_orders[ctx->fset][i])
1356 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
1357 ape_filter_orders[ctx->fset][i],
1358 ape_filter_fracbits[ctx->fset][i]);
1362 static int init_frame_decoder(APEContext *ctx)
1365 if ((ret = init_entropy_decoder(ctx)) < 0)
1367 init_predictor_decoder(ctx);
1369 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1370 if (!ape_filter_orders[ctx->fset][i])
1372 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
1373 ape_filter_orders[ctx->fset][i]);
1378 static void ape_unpack_mono(APEContext *ctx, int count)
1380 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1381 /* We are pure silence, so we're done. */
1382 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
1386 ctx->entropy_decode_mono(ctx, count);
1390 /* Now apply the predictor decoding */
1391 ctx->predictor_decode_mono(ctx, count);
1393 /* Pseudo-stereo - just copy left channel to right channel */
1394 if (ctx->channels == 2) {
1395 memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
1399 static void ape_unpack_stereo(APEContext *ctx, int count)
1401 unsigned left, right;
1402 int32_t *decoded0 = ctx->decoded[0];
1403 int32_t *decoded1 = ctx->decoded[1];
1405 if ((ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) == APE_FRAMECODE_STEREO_SILENCE) {
1406 /* We are pure silence, so we're done. */
1407 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
1411 ctx->entropy_decode_stereo(ctx, count);
1415 /* Now apply the predictor decoding */
1416 ctx->predictor_decode_stereo(ctx, count);
1418 /* Decorrelate and scale to output depth */
1420 left = *decoded1 - (unsigned)(*decoded0 / 2);
1421 right = left + *decoded0;
1423 *(decoded0++) = left;
1424 *(decoded1++) = right;
1428 static int ape_decode_frame(AVCodecContext *avctx, void *data,
1429 int *got_frame_ptr, AVPacket *avpkt)
1431 AVFrame *frame = data;
1432 const uint8_t *buf = avpkt->data;
1433 APEContext *s = avctx->priv_data;
1439 uint64_t decoded_buffer_size;
1441 /* this should never be negative, but bad things will happen if it is, so
1442 check it just to make sure. */
1443 av_assert0(s->samples >= 0);
1446 uint32_t nblocks, offset;
1453 if (avpkt->size < 8) {
1454 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1455 return AVERROR_INVALIDDATA;
1457 buf_size = avpkt->size & ~3;
1458 if (buf_size != avpkt->size) {
1459 av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
1460 "extra bytes at the end will be skipped.\n");
1462 if (s->fileversion < 3950) // previous versions overread two bytes
1464 av_fast_padded_malloc(&s->data, &s->data_size, buf_size);
1466 return AVERROR(ENOMEM);
1467 s->bdsp.bswap_buf((uint32_t *) s->data, (const uint32_t *) buf,
1469 memset(s->data + (buf_size & ~3), 0, buf_size & 3);
1471 s->data_end = s->data + buf_size;
1473 nblocks = bytestream_get_be32(&s->ptr);
1474 offset = bytestream_get_be32(&s->ptr);
1475 if (s->fileversion >= 3900) {
1477 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
1480 return AVERROR_INVALIDDATA;
1482 if (s->data_end - s->ptr < offset) {
1483 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1484 return AVERROR_INVALIDDATA;
1488 if ((ret = init_get_bits8(&s->gb, s->ptr, s->data_end - s->ptr)) < 0)
1490 if (s->fileversion > 3800)
1491 skip_bits_long(&s->gb, offset * 8);
1493 skip_bits_long(&s->gb, offset);
1496 if (!nblocks || nblocks > INT_MAX / 2 / sizeof(*s->decoded_buffer) - 8) {
1497 av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %"PRIu32".\n",
1499 return AVERROR_INVALIDDATA;
1502 /* Initialize the frame decoder */
1503 if (init_frame_decoder(s) < 0) {
1504 av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
1505 return AVERROR_INVALIDDATA;
1507 s->samples = nblocks;
1515 blockstodecode = FFMIN(s->blocks_per_loop, s->samples);
1516 // for old files coefficients were not interleaved,
1517 // so we need to decode all of them at once
1518 if (s->fileversion < 3930)
1519 blockstodecode = s->samples;
1521 /* reallocate decoded sample buffer if needed */
1522 decoded_buffer_size = 2LL * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer);
1523 av_assert0(decoded_buffer_size <= INT_MAX);
1525 /* get output buffer */
1526 frame->nb_samples = blockstodecode;
1527 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) {
1532 av_fast_malloc(&s->decoded_buffer, &s->decoded_size, decoded_buffer_size);
1533 if (!s->decoded_buffer)
1534 return AVERROR(ENOMEM);
1535 memset(s->decoded_buffer, 0, decoded_buffer_size);
1536 s->decoded[0] = s->decoded_buffer;
1537 s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);
1541 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
1542 ape_unpack_mono(s, blockstodecode);
1544 ape_unpack_stereo(s, blockstodecode);
1549 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
1550 return AVERROR_INVALIDDATA;
1555 for (ch = 0; ch < s->channels; ch++) {
1556 sample8 = (uint8_t *)frame->data[ch];
1557 for (i = 0; i < blockstodecode; i++)
1558 *sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
1562 for (ch = 0; ch < s->channels; ch++) {
1563 sample16 = (int16_t *)frame->data[ch];
1564 for (i = 0; i < blockstodecode; i++)
1565 *sample16++ = s->decoded[ch][i];
1569 for (ch = 0; ch < s->channels; ch++) {
1570 sample24 = (int32_t *)frame->data[ch];
1571 for (i = 0; i < blockstodecode; i++)
1572 *sample24++ = s->decoded[ch][i] * 256U;
1577 s->samples -= blockstodecode;
1579 if (avctx->err_recognition & AV_EF_CRCCHECK &&
1580 s->fileversion >= 3900 && s->bps < 24) {
1581 uint32_t crc = s->CRC_state;
1582 const AVCRC *crc_tab = av_crc_get_table(AV_CRC_32_IEEE_LE);
1583 for (i = 0; i < blockstodecode; i++) {
1584 for (ch = 0; ch < s->channels; ch++) {
1585 uint8_t *smp = frame->data[ch] + (i*(s->bps >> 3));
1586 crc = av_crc(crc_tab, crc, smp, s->bps >> 3);
1590 if (!s->samples && (~crc >> 1) ^ s->CRC) {
1591 av_log(avctx, AV_LOG_ERROR, "CRC mismatch! Previously decoded "
1592 "frames may have been affected as well.\n");
1593 if (avctx->err_recognition & AV_EF_EXPLODE)
1594 return AVERROR_INVALIDDATA;
1602 return !s->samples ? avpkt->size : 0;
1605 static void ape_flush(AVCodecContext *avctx)
1607 APEContext *s = avctx->priv_data;
1611 #define OFFSET(x) offsetof(APEContext, x)
1612 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1613 static const AVOption options[] = {
1614 { "max_samples", "maximum number of samples decoded per call", OFFSET(blocks_per_loop), AV_OPT_TYPE_INT, { .i64 = 4608 }, 1, INT_MAX, PAR, "max_samples" },
1615 { "all", "no maximum. decode all samples for each packet at once", 0, AV_OPT_TYPE_CONST, { .i64 = INT_MAX }, INT_MIN, INT_MAX, PAR, "max_samples" },
1619 static const AVClass ape_decoder_class = {
1620 .class_name = "APE decoder",
1621 .item_name = av_default_item_name,
1623 .version = LIBAVUTIL_VERSION_INT,
1626 AVCodec ff_ape_decoder = {
1628 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
1629 .type = AVMEDIA_TYPE_AUDIO,
1630 .id = AV_CODEC_ID_APE,
1631 .priv_data_size = sizeof(APEContext),
1632 .init = ape_decode_init,
1633 .close = ape_decode_close,
1634 .decode = ape_decode_frame,
1635 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DELAY |
1637 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
1639 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
1642 AV_SAMPLE_FMT_NONE },
1643 .priv_class = &ape_decoder_class,